In [108]:
# !pip install seaborn
import ipywidgets as widgets
from ipywidgets import interact, interact_manual
In [13]:
# ! pip install --upgrade seaborn
In [3]:
import pandas as pd
In [4]:
wine_data = pd.read_csv('winequality-white.csv')
In [5]:
wine_data.head()
Out[5]:
In [6]:
wine_data.columns
Out[6]:
In [7]:
wine_data.tail()
Out[7]:
In [8]:
len(wine_data)
Out[8]:
In [9]:
wine_data.describe()
Out[9]:
distplot¶
In [14]:
import seaborn as sns
import matplotlib.pyplot as plt
In [15]:
wine_data = pd.read_csv('winequality-white.csv')
wine_data.columns = ['fixed_acidity', 'volatile_acidity', 'citric_acid', 'residual_sugar',
'chlorides', 'free_sulfur_dioxide', 'total_sulfur_dioxide', 'density',
'pH', 'sulphates', 'alcohol', 'quality']
In [16]:
wine_data.tail(10)
Out[16]:
In [17]:
plt.hist(wine_data.alcohol)
Out[17]:
In [18]:
sns.distplot(wine_data.alcohol, kde=True) # histogram bars KDE:gaussian kernel density estimate.
Out[18]:
around 40% have alcohol content = 9.3
In [19]:
f,ax = plt.subplots(figsize = (15,5))
sns.distplot(wine_data.alcohol, kde=True)
Out[19]:
In [20]:
f,ax = plt.subplots(figsize = (15,5))
sns.distplot(wine_data.alcohol, kde=False, rug=True , bins = 200)
Out[20]:
rug: Whether to draw a rugplot on the support axis.
bins control granularity of the bars , bins = more size -> you can analyse the data more deep
In [21]:
f,ax = plt.subplots(figsize = (15,5))
sns.rugplot(wine_data.alcohol, height = 0.75) # most of the rug distribution is clustered around centre
Out[21]:
In [22]:
f,ax = plt.subplots(figsize = (15,5))
sns.distplot(wine_data.alcohol, kde=True, rug=True, hist = False)
Out[22]:
shade the Kernal density estimation area¶
In [23]:
f,ax = plt.subplots(figsize = (15,5))
sns.set(color_codes=True)
sns.kdeplot(wine_data.alcohol, shade=True, color ='r')
Out[23]:
plotting KDE with various bandwidth that fits the original data¶
In [24]:
f,ax = plt.subplots(figsize=(15,5))
sns.kdeplot(wine_data.alcohol)
sns.kdeplot(wine_data.alcohol, bw = 0.04 , label = 'bw =0.04 ' )
sns.kdeplot(wine_data.alcohol, bw = 0.2 , label = 'bw =0.2 ' )
sns.kdeplot(wine_data.alcohol, bw = 2 , label = 'bw =2 ' )
sns.kdeplot(wine_data.alcohol, bw = 5 , label = 'bw =5 ' )
Out[24]:
the more erratic the plot is the less generic information it provides¶
Joint Plots along with univariate distribution¶
In [25]:
sns.jointplot(x = wine_data.free_sulfur_dioxide, y= wine_data.total_sulfur_dioxide, xlim={0,150}, ylim={0,400}, alpha=0.2) # ALPHA STILL WORKS HERE
Out[25]:
Hexbin Plots - Histogram representation of bivariate plot¶
the problem here in jointplot is at the middle we can not decide the relationship between wine_data.free_sulfur_dioxide wine_data.total_sulfur_dioxide ; solution to this is plotting Hexbin plot with Hue variations¶
how dark and how light a perticular point is represents the height of histogram at intersection¶
In [26]:
sns.jointplot(x = wine_data.free_sulfur_dioxide, y= wine_data.total_sulfur_dioxide, kind='hex', xlim={0,100}, ylim={0,200})
Out[26]:
KDE curves for bivariate distribution¶
In [27]:
sns.jointplot(x = wine_data.free_sulfur_dioxide, y= wine_data.total_sulfur_dioxide, kind='kde', xlim={0,80}, ylim={0,300})
Out[27]:
In [28]:
f,ax = plt.subplots(figsize = (8,5))
sns.kdeplot(wine_data.free_sulfur_dioxide, wine_data.total_sulfur_dioxide)
sns.rugplot(wine_data.free_sulfur_dioxide, color = 'g',height=0.05)
sns.rugplot(wine_data.total_sulfur_dioxide, color = 'b', height=0.05, vertical = True)
plt.xlim(-20,100)
plt.ylim(-50,400)
Out[28]:
plot bivariate relationships between every pair of columns in a dataset¶
In [29]:
# sns.pairplot(wine_data, height=3) # rn this it takes a lot of time
In [30]:
sns.pairplot(wine_data, height=3, vars=['fixed_acidity','chlorides','sulphates', 'alcohol', 'quality'], diag_kind='kde')
Out[30]:
pairwise regressions¶
In [31]:
sns.pairplot(wine_data, height=3, vars=['chlorides','sulphates', 'quality'], kind='reg')
Out[31]:
In [32]:
g= sns.PairGrid(wine_data, height=3, vars=['chlorides','sulphates', 'quality'])
g.map(plt.scatter)
Out[32]:
dont you think it is stupid to plot a univariate varaible against itself... ok lets do one thing let's plot KDE for diagonal and scatter for non diagonal plots¶
In [33]:
g= sns.PairGrid(wine_data,vars=['chlorides','sulphates', 'alcohol'], hue='quality')
g.map_offdiag(plt.scatter)
g.map_diag(sns.kdeplot)
plt.legend(loc=2)
Out[33]:
Now how about tuning PairGrid even further map_upper map_lower map_diag¶
In [34]:
g=sns.PairGrid(wine_data,vars=['chlorides','sulphates', 'alcohol'])
g.map_diag(sns.kdeplot)
g.map_upper(sns.scatterplot)
g.map_lower(sns.regplot)
Out[34]:
Howabout using PairGrid with x_vars and y_vars differet¶
In [35]:
g=sns.PairGrid(wine_data,x_vars=['chlorides','sulphates', 'fixed_acidity'], y_vars=['alcohol'])
g.map(plt.scatter)
# g.map_diag(sns.kdeplot)
# g.map_upper(sns.scatterplot)
# g.map_lower(sns.regplot)
Out[35]:
In [36]:
g=sns.PairGrid(wine_data,x_vars=['chlorides','sulphates', 'fixed_acidity'], y_vars=['alcohol'])
g.map(sns.scatterplot) # sns looks more beautiful
Out[36]:
In [37]:
# sns.pairplot(wine_data, height=3, vars=['fixed_acidity','chlorides','sulphates', 'quality'], hue = 'pH')
Heatmaps¶
In [38]:
corrmat = wine_data.corr()
f,ax = plt.subplots(figsize=(10,10))
# sns.heatmap(corrmat, vmin = -0.9, vmax=0.95, square = True, annot= True, fmt='.2f', cmap='summer' )
sns.heatmap(corrmat, annot=True, fmt='0.2f', square = True)
Out[38]:
lmplots and regplots for regression¶
In [39]:
sns.lmplot(x='residual_sugar', y='density', data = wine_data, height = 7, aspect=2)
Out[39]:
when MSE is low CI is narrow and whereever the MSE is high CI-confidence Interval is High¶
categorical variable with discrete values¶
In [40]:
sns.lmplot(x='quality', y='alcohol', data=wine_data)
Out[40]:
hard to see individual data points so add some jittering - regression line is not affected by jittering¶
In [41]:
sns.lmplot(x='quality', y='alcohol', data=wine_data, x_jitter=.2)
Out[41]:
In [42]:
import numpy as np
sns.lmplot(x='quality', y='alcohol', data=wine_data, x_estimator=np.mean )
Out[42]:
Higher quality means higher alcohol content¶
In [43]:
# sns.lmplot(x='pH',y='fixed_acidity', data= wine_data, row='quality', hue='alcohol') # Fixed_acidity vs pH in each quality category with hue= alcohol : needs tuning
In [44]:
# sns.lmplot(x='pH',y='fixed_acidity', data= wine_data, col='quality', hue='alcohol') # Fixed_acidity vs pH in each quality category with hue= alcohol
regplot¶
In [45]:
sns.regplot(wine_data.alcohol, wine_data.density, color ='0')
Out[45]:
In [46]:
f, ax = plt.subplots(figsize=(12,5))
sns.regplot(x=wine_data.residual_sugar, y=wine_data.density, ax=ax) # controlling size and shape
Out[46]:
joint plots also plots regression if kind is specified as 'reg'¶
In [47]:
sns.jointplot(x = wine_data.free_sulfur_dioxide, y= wine_data.total_sulfur_dioxide, xlim={0,150}, ylim={0,400}, kind='reg')
Out[47]:
In [48]:
sns.pairplot(wine_data,x_vars=['fixed_acidity', 'citric_acid', 'chlorides'], y_vars=['alcohol'], kind='reg', height=8, aspect=1)
Out[48]:
CATEGORICAL AND MULTIPANEL DATA¶
In [49]:
f,ax=plt.subplots(figsize=(15,5))
sns.stripplot(x='quality',y='alcohol',data= wine_data)
Out[49]:
In [50]:
f,ax=plt.subplots(figsize=(15,5))
sns.swarmplot(x='quality',y='alcohol',data= wine_data, hue='pH')
Out[50]:
In [51]:
sns.dogplot()
BOX PLOT¶
In [52]:
f,ax = plt.subplots(figsize=(15,4))
sns.boxplot(x='quality', y='alcohol', data=wine_data)
Out[52]:
wiskers are mostly 1.5 times inter quantile ; any data outside this range is represented as outlier seperately¶
VIOLIN PLOT mit distribution probability¶
In [53]:
f,ax = plt.subplots(figsize=(15,4))
sns.violinplot(x='quality', y='alcohol', data=wine_data)
Out[53]:
The curved boundaries represents KDE , violin plot also shows mode ; violin plots can also be plotted relative to counts of individual data¶
In [54]:
f,ax = plt.subplots(figsize=(15,4))
sns.violinplot(x='quality', y='alcohol', data=wine_data, scale='count')
Out[54]:
here you can scale the width with respect to number of counts of each categorical observation¶
In [55]:
f,ax = plt.subplots(figsize=(15,4))
sns.violinplot(x='quality', y='alcohol', data=wine_data, scale='count', inner='stick')
Out[55]:
scaled thickly closed lines suggest there are many samples in our data with quality 6 and at alcohol level 11.5¶
COMBINING SWARM AND VIOLIN PLOTS¶
In [56]:
f,ax = plt.subplots(figsize=(15,4))
sns.violinplot(x='quality', y='alcohol', data=wine_data)
sns.swarmplot(x='quality',y='alcohol',data= wine_data)
Out[56]:
STATISTICAL ESTIMATION¶
In [57]:
f, ax= plt.subplots(figsize=(15,5))
sns.barplot(x='quality', y='pH', data=wine_data)
Out[57]:
In [58]:
f, ax= plt.subplots(figsize=(15,5))
sns.countplot(x='alcohol', data=wine_data)
Out[58]:
In [59]:
f, ax= plt.subplots(figsize=(20,5))
sns.countplot(x='alcohol', data=wine_data, color = 'm', palette='Greens_d')
plt.xticks(rotation=90)
Out[59]:
Number of wines at every alcohollevel¶
POINT PLOT¶
A point plot represents an estimate of central tendency for a numeric variable by the position of scatter plot points and provides some indication of the uncertainty around that estimate using error bars.¶
In [60]:
f, ax= plt.subplots(figsize=(20,5))
sns.pointplot(x='quality', y='pH', data=wine_data)
Out[60]:
In [62]:
# f, ax= plt.subplots(figsize=(28,5))
# sns.pointplot(x='quality', y='pH', data=wine_data, hue='alcohol') # hue here helps us to visualise the variation of the
Observe how the alcohol content varies with the ph and quality of wine¶
BOX PLOT¶
In [63]:
f,ax = plt.subplots(figsize=(15,8))
sns.boxplot(data= wine_data, orient='h')
Out[63]:
Seaborn is a powerful visualization library built on top of matplotlib ; tightly integrated with PyData stack and makes production ready plots¶
Facet grids in Seaborn¶
In [65]:
titanic = pd.read_csv('titanic.csv')
In [66]:
titanic.head()
Out[66]:
In [67]:
g = sns.FacetGrid(titanic, col='Pclass')
In [68]:
g = sns.FacetGrid(titanic, col='Survived')
g.map(plt.hist,'Pclass', color='k')
Out[68]:
Plotting bivariate data¶
In [69]:
g = sns.FacetGrid(titanic, col='Survived', height=8)
g.map(plt.scatter,'Age','Fare')
Out[69]:
In [70]:
g = sns.FacetGrid(titanic, col='Survived', height=8, hue='Sex')
g.map(plt.scatter,'Age','Fare', alpha= 0.99) # alpha is the opacity
g.add_legend()
Out[70]:
In [71]:
g = sns.FacetGrid(titanic, col='Survived', height=8, hue='Sex')
g.map(plt.bar,'Age','Fare', alpha= 0.99) # alpha is the opacity
g.add_legend()
Out[71]:
In [72]:
g = sns.FacetGrid(titanic, col='Survived', row = 'Pclass', height=8, hue='Sex')
g.map(plt.bar,'Age','Fare', alpha= 0.99) # alpha is the opacity
g.add_legend()
Out[72]:
In [73]:
g = sns.FacetGrid(titanic, col='Survived', row = 'Pclass', height=8, hue='Sex')
g.map(sns.barplot,'Age','Fare')
Out[73]:
In [74]:
# g = sns.FacetGrid(titanic, col='Survived', row = 'Pclass', height=8, hue='Sex')
# g.map(sns.regplot,'Age','Fare', fit_reg=False) # alpha is the opacity
# g.add_legend()
In [75]:
g = sns.FacetGrid(titanic, col='Survived', row = 'Pclass', height=8, hue='Sex')
g.map(plt.scatter,'Age','Fare', alpha= 0.99) # alpha is the opacity
g.add_legend()
Out[75]:
In [76]:
# How about changing some colors
h={"male":'b',"female":'r'}
g = sns.FacetGrid(titanic, col='Survived', row = 'Pclass', height=8, hue='Sex', palette=h)
g.map(plt.scatter,'Age','Fare', alpha= 0.99) # alpha is the opacity
g.add_legend()
Out[76]:
NOW HOW ABOUT SOME FINE TUNIG!!!¶
In [77]:
g = sns.FacetGrid(titanic, col='Survived', height=8, col_wrap = 4)
g.map(sns.barplot,'Age','Fare')
Out[77]:
In [78]:
g = sns.FacetGrid(titanic, col='Survived', row = 'Pclass', height=8, hue='Sex')
g.map(sns.barplot,'Age','Fare', color='#334488', edgecolor ='red',lw=.5 ) # color is a hex parameter
plt.xticks(rotation=90)
g.fig.subplots_adjust(wspace = 0.3, hspace=0.5) # giving some white space and horizontal spacing
Changing labels¶
In [79]:
g = sns.FacetGrid(titanic, col='Survived', row = 'Pclass', height=8)
g.map(sns.barplot,'Age','Fare', color='#334488', edgecolor ='red',lw=.5 ) # color is a hex parameter
plt.xticks(rotation=90)
g.fig.subplots_adjust(wspace = 0.3, hspace=0.5) # giving some white space and horizontal spacing
g.set_axis_labels('Age of Passengers','Fare of each Passenger')
Out[79]:
is this bar graph showing frequency of fares? no !! it is just giving one farevalue for each bucket of age; work on it later now!¶
Customizing the yticks¶
In [80]:
g = sns.FacetGrid(titanic, col='Survived', height=8, hue='Sex')
g.map(sns.scatterplot,'Age','Fare') # color is a hex parameter
g.fig.subplots_adjust(wspace = 0.3, hspace=0.5) # giving some white space and horizontal spacing
g.set_axis_labels('Age of Passengers','Fare of each Passenger')
g.set(yticks=[0,50,100,150,200,250,300,350,400,450,500])
Out[80]:
xlim and ylim - limit your x and y ticks¶
In [81]:
g = sns.FacetGrid(titanic, col='Survived', height=5, hue='Sex', xlim=(0,16), ylim=(100,500))
g.map(sns.scatterplot,'Age','Fare') # color is a hex parameter
g.fig.subplots_adjust(wspace = 0.3, hspace=0.5) # giving some white space and horizontal spacing
g.set_axis_labels('Age of Passengers','Fare of each Passenger')
Out[81]:
this is function inbulit argument method to restrict our x and y ticks¶
In [82]:
g = sns.FacetGrid(titanic, col='Survived', height=5, hue='Sex')
g.map(sns.scatterplot,'Age','Fare') # color is a hex parameter
g.fig.subplots_adjust(wspace = 0.3, hspace=0.5) # giving some white space and horizontal spacing
g.set_axis_labels('Age of Passengers','Fare of each Passenger')
g.set(xlim=(0,16), ylim=(100,500))
Out[82]:
Machine learning bicycle data set¶
In [83]:
bike = pd.read_csv('bike_sharing_daily.csv')
In [84]:
bike.head()
Out[84]:
In [85]:
bike.columns
Out[85]:
In [86]:
bike.columns=['instant', 'date', 'season', 'year', 'month', 'holiday', 'weekday',
'workingday', 'weathersituation', 'temp', 'atemp', 'humidity', 'windspeed',
'casual_users', 'registered_users', 'cnt']
In [87]:
# bike.cnt.head(200) bike count per day may go above 8000 per day and some days as low as 100 bikes per day
In [88]:
f,ax = plt.subplots(figsize=(15,5))
sns.set(style="whitegrid")
# sns.set_xticklabels=([0,500,1000,1500,2000,2500,3000,3500,4000,4500,5000,5500,6000,6500,7000,7500,8000,8500,9000])
sns.distplot(bike.cnt, bins=120, kde=False, rug=True, color='k')
Out[88]:
In [89]:
f,ax = plt.subplots(figsize=(15,5))
sns.set(style="whitegrid")
# sns.set_xticklabels=([0,500,1000,1500,2000,2500,3000,3500,4000,4500,5000,5500,6000,6500,7000,7500,8000,8500,9000])
sns.distplot(bike.cnt, bins=120, kde=True, rug=True, color='k')
Out[89]:
In [90]:
sns.jointplot(bike.instant,bike.cnt, height=8, color='g') # number of bikes hired as a function of day passed
Out[90]:
two peaks in two years¶
In [91]:
sns.jointplot(bike.month,bike.cnt, height=8, color='g')
Out[91]:
monthwise counts; as seen in months 8,9,10 more bicycles are rented¶
season wise rental data¶
In [92]:
sns.jointplot(bike.season,bike.cnt) # counts are higher in season 3 and 4
Out[92]:
In [93]:
sns.pairplot(bike,height=6, aspect=1.2,x_vars=['temp', 'humidity', 'windspeed'], y_vars='cnt', hue='season')
Out[93]:
Themes and Styles¶
In [95]:
# sns.set_style('darkgrid') # applies to all lines ; affects all plots that follow
# f,ax = plt.subplots(figsize=(15,5))
# sns.scatterplot(bike.cnt)
In [96]:
with sns.axes_style('darkgrid'): # applies to only this lines
f,ax = plt.subplots(figsize=(15,5))
sns.distplot(bike.cnt)
How to remove spines¶
In [97]:
sns.jointplot(bike.instant,bike.cnt, height=8, color='g')
sns.despine(offset=15, trim=True)
In [98]:
sns.set()
In [99]:
sns.jointplot(bike.instant,bike.cnt, height=8, color='g')
Out[99]:
colors¶
In [100]:
current_palette = sns.color_palette()
sns.palplot(current_palette)
In [101]:
sns.palplot(sns.color_palette('hls',8)) # hls-> Hue Lightness and Saturation
In [102]:
sns.palplot(sns.hls_palette(n_colors=15, h=.99, l=0.5, s=.99))
In [105]:
sns.choose_colorbrewer_palette()
In [109]:
# !pip install ipywidgets
In [111]:
sns.choose_colorbrewer_palette('diverging')
In [112]:
# sequential color plots ;
sns.palplot(sns.color_palette('Blues'))
In [113]:
# sequential color plots ;
sns.palplot(sns.color_palette('Blues_d'))
In [114]:
sns.choose_colorbrewer_palette('sequential')
In [115]:
sns.palplot(sns.color_palette('cubehelix',12))
In [116]:
sns.palplot(sns.cubehelix_palette(12)) # good for printing
Diverging color palette¶
In [117]:
h = sns.choose_colorbrewer_palette('diverging')
In [118]:
h = sns.choose_colorbrewer_palette('sequential')
g=sns.PairGrid(wine_data,x_vars=['chlorides','sulphates', 'fixed_acidity'], y_vars=['alcohol'], hue = 'pH',palette=h, height= 8)
g.map(sns.scatterplot) # sns looks more beautiful
# g.add_legend()
Aesthetics¶
In [119]:
sns.axes_style() # current style
Out[119]:
In [120]:
sns.set_style('ticks',{'xtick.major.size':8,'xtick.color': '.15','ytick.color': '.85','ytick.major.size':10,'axes.facecolor': 'm'})
sns.jointplot(bike.instant,bike.cnt, height=8, color='k')
Out[120]:
In [121]:
sns.set() # reset everything
In [122]:
sns.jointplot(bike.instant,bike.cnt, height=8, color='k')
Out[122]:
printable and seminar contexts¶
In [123]:
sns.set_context('talk')
sns.jointplot(bike.instant,bike.cnt, height=8, color='k')
Out[123]:
In [124]:
sns.set_context('paper')
sns.jointplot(bike.instant,bike.cnt, height=8, color='k')
Out[124]:
In [125]:
sns.set_context('notebook')
sns.jointplot(bike.instant,bike.cnt, height=8, color='k')
Out[125]:
In [126]:
sns.set_context('poster')
sns.jointplot(bike.instant,bike.cnt, height=8, color='k')
Out[126]:
continued!!!!!¶
In [ ]: